Probabilistic just-in-time approach for nonlinear modeling with Bayesian nonlinear feature extraction

Shen, Feifeng; Jan, Nabil Magbool; Huang, Biao; Yang, Huizhong

doi:10.1016/j.chemolab.2019.103895

Cited by 6 publications

(6 citation statements)

References 20 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…JIT modelling has found several applications in soft sensing. [19][20][21] For the mining application under consideration, the geological block model can be thought of as a global model since, once it is built, it is used for resource or mine planning and for predicting ore characteristics. Furthermore, it should be noted that the geological block model development is a challenging and time consuming task, so that a yearly update of the model is not desired and a shorter time frame would be unwarranted.…”

Section: Just-in-time Modelling Frameworkmentioning

confidence: 99%

Just‐in‐time learning for the prediction of oil sands ore characteristics using GPS data in mining applications

et al. 2020

Self Cite

View full text Add to dashboard Cite

For the mining based oilsands industry, it is desirable to determine the quality of the ore delivered to the extraction processes in real‐time to make optimal operational decisions such as optimal ore blending to achieve maximal bitumen recovery. Currently, the industry determines the real‐time ore characteristics for any given shovel Global Positioning System (GPS) location by first determining the shovel elevation from the topological mine map and then using the determined geological coordinates in the 3D geological block model. It should be noted that the block model is built based on the widely spaced core hole samples, and it is updated only on a yearly basis due to high cost of narrower core hole sampling. Thus, the block model predictions are often inaccurate in between the core hole spacing. On the other hand, mining operations data are available that contain accurate ore characteristics information in the already mined area. Therefore, in this work, we present a just‐in‐time based data‐driven modelling strategy that utilizes the recently available mining operations data to obtain reliable ore characteristics given the GPS data. The prediction capability of ore characteristics using the proposed modelling strategy is validated at core hole locations. Further, the prediction of ore characteristics at non‐core hole points demonstrate promising results.

show abstract

Section: Just-in-time Modelling Frameworkmentioning

confidence: 99%

Just‐in‐time learning for the prediction of oil sands ore characteristics using GPS data in mining applications

et al. 2020

Self Cite

View full text Add to dashboard Cite

show abstract

“…Therefore, developing local models would be a reasonable solution to the approximation of complex processes . According to the state-of-art research, there are several local model methods that can be considered. − Some mechanism models that are useful to local models, such as the neuro-fuzzy network and the Takagi–Sugeno (T–S) fuzzy model, require prior knowledge to determine the operating space, model structure, and parameters. Nevertheless, when dealing with fairly complex or even unknown processes, these methods have shortcomings (such as huge workload of constructing models and becoming unusable to unknown processes).…”

Section: Introductionmentioning

confidence: 99%

“…Nevertheless, when dealing with fairly complex or even unknown processes, these methods have shortcomings (such as huge workload of constructing models and becoming unusable to unknown processes). Conversely, just-in-time learning (JITL) does not require mechanism knowledge but uses data-driven patterns to generate local models. ,− JITL uses a designed similarity criterion to select the most relevant historical samples for the constructions of local models. In particular, the local models are discarded immediately after predicting the output value of the query samples.…”

Section: Introductionmentioning

confidence: 99%

“…However, constructing local models by PJITL is not flexible enough. In terms of the size of local models, all the local models, using PJITL or other just-in-time learning methods, share the same size which is determined by minimizing prediction errors. ,, It might lead to several situations: First, when the local model is in a suitable size, the query sample could get a better prediction and only needs relatively less time to calculate. Second, if the size of the local model is small for the query sample, we could not get enough information from the limited training samples so that the prediction would be unreliable.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Variable-Scale Probabilistic Just-in-Time Learning for Soft Sensor Development with Missing Data

Huang

Peng

Jiang

et al. 2020

Ind. Eng. Chem. Res.

View full text Add to dashboard Cite

Just-in-time learning (JITL) has been widely applied to data-driven modeling to deal with the nonlinearity problems in industrial processes. To mitigate the effects of noise existing in JITL, probabilistic JITL (PJITL) selects samples based on the probability distributions. Considering the existence of missing data situation, the PJITL algorithm could also cope with that. However, traditional JITL-based methods, including PJITL, cannot flexibly select the number of training samples for each query sample, which would in return influence the accuracy of prediction for a part of query samples. To solve this problem, we proposed a method named “variable-scale PJITL” (VS-PJITL) which can determine the sizes of the local models for each query sample using a new sample selection criterion. Based on the Euclidean distance, the sample selection criterion also applies to the variable-scale JITL (VS-JITL). Then, comparisons of VS-PJITL, PJITL, JITL, and VS-JITL are tested on a simulated data set and a real industrial data set from the catalytic naphtha reforming process. By analyzing the two cases above, VS-PJITL is considered to have superior performance to the original PJITL (root-mean-square error reduced by 0.3355 and 0.4778).

show abstract

“…proposed a multisimilarity measure including distance-based similarity criteria and nondistance-based similarity criteria, which addressed the limitation when using one similarity criterion. Besides, some similarity measures based on the Kullback–Leibler (KL) divergence have been put forward recently. , However, some drawbacks of the aforementioned similarity criteria reduce the accuracy of modeling data selection in JITL-based soft sensor for batch processes: (1) the modeling data search range of these similarity criteria is the entire batch, which neglects the multiphase characteristic of batch process data, and (2) these similarity criteria only consider the similarity between the input data. For soft sensor model training, the similarity between the output data is also important.…”

Section: Introductionmentioning

confidence: 99%

Soft Sensor Framework Based on Semisupervised Just-in-Time Relevance Vector Regression for Multiphase Batch Processes with Unlabeled Data

Qiu

Wang

Zhou

et al. 2020

Ind. Eng. Chem. Res.

View full text Add to dashboard Cite

Soft sensors using just-in-time learning (JITL) have attracted much attention in the application of online prediction in batch processes because of the ability to perform adaptive updating and dynamic modeling. However, developing effective JITL-based soft sensors of batch processes remains challenging due to the unlabeled data caused by the expensive online measuring instruments and long time-consuming offline analysis. Besides, the multiphase and nonlinear characteristics of batch processes make this challenge more complicated. To cope with this challenge, a novel soft sensor framework, termed semisupervised just-in-time relevance vector regression (SJRVR), is proposed. The SJRVR integrates JITL, adversarial autoencoder (AAE), and RVR into a unified framework to address soft sensor modeling for multiphase batch processes with unlabeled data. In this framework, an unlabeled data processing strategy based on process mechanism and AAE (MAAE) is presented to utilize the useful information on unlabeled data. Moreover, a local regression model is constructed using JITL with a designed modeling data selection strategy and RVR to address the soft sensor modeling for process data with multiphase and nonlinear characteristics. The effectiveness of the SJRVR method is illustrated with a penicillin fermentation process.

show abstract

Probabilistic just-in-time approach for nonlinear modeling with Bayesian nonlinear feature extraction

Cited by 6 publications

References 20 publications

Just‐in‐time learning for the prediction of oil sands ore characteristics using GPS data in mining applications

Just‐in‐time learning for the prediction of oil sands ore characteristics using GPS data in mining applications

Variable-Scale Probabilistic Just-in-Time Learning for Soft Sensor Development with Missing Data

Soft Sensor Framework Based on Semisupervised Just-in-Time Relevance Vector Regression for Multiphase Batch Processes with Unlabeled Data

Contact Info

Product

Resources

About